Computer-implemented method and system for testing a model

ChangeBridge: Patent Grants - Business Methods (G06Q)

Published March 31st, 2026

Detected March 31st, 2026

Summary

The USPTO granted Mastercard International Inc. Patent No. US12591836B2 for a computer-implemented method of testing AI/ML models using statistical p-value analysis. The method compares distributions of p-values from real datasets against control datasets to determine model accuracy for multivariate data analysis.

View original document View source feed page

What changed

Mastercard International Inc. has been granted US Patent 12,591,836 covering a method and system for testing models indicating parametric relationships in multivariate data. The patent describes generating control datasets based on the model, calculating p-values for sample subsets from both real and control datasets, and determining model accuracy by comparing distribution characteristics against predetermined significance thresholds.

This is a patent grant announcement, not a regulatory requirement. No compliance actions, deadlines, or penalties apply. Companies developing AI/ML models for financial applications may reference this patent for technical methodology insights, but no immediate regulatory response is required.

Source document (simplified)

← USPTO Patent Grants

Computer-implemented method and system for testing a model

Grant US12591836B2 Kind: B2 Mar 31, 2026

Assignee

MASTERCARD INTERNATIONAL INCORPORATED

Inventors

James Conway, Quentin Bragard

Abstract

A computer-implemented method for testing a model indicating a parametric relationship between a plurality of channels of multivariate data, the method comprising: obtaining a real dataset of observed multivariate data comprising the plurality of channels; generating a control dataset of multivariate data comprising the plurality of channels, the control dataset being generated based on the model; for each of a plurality of sample subsets from the real dataset and the control dataset, calculating a p-value for the model; and determining whether (1) a distribution difference characteristic of a distribution of the obtained p-values for sample subsets of the real dataset and a distribution of the obtained p-values for sample subsets of the control dataset falls within a second predetermined significance range, and, if condition (1) is met, determining that the model is accurate.